Artem Sevastopolsky

how to pronounce my last name: [see vast awe pole ski]

I am a Ph.D. student at TUM (Visual Computing & AI lab led by prof. Matthias Nießner). Previously, I worked as a deep learning engineer at the Vision, Learning and Telepresence lab at Samsung AI Center Moscow and studied at Skoltech under supervision of prof. Victor Lempitsky. Currently, I am a research intern at Meta Reality Labs, Codec Avatars Team.

At Samsung & Skoltech I worked on realistic differentiable rendering of 3D point clouds, human photo-to-texture synthesis, differentiable warping methods for view resynthesis and other projects. Previously, I was more into medical imaging at Lomonosov Moscow State University and in the industry.

Email  /  CV  /  Google Scholar  /  GitHub  /  LinkedIn  /  Twitter

profile photo
Featured Research

At the moment I'm mostly interested in applications of computer vision to 3D computer graphics and rendering.

HeadCraft: Modeling High-Detail Shape Variations for Animated 3DMMs
Artem Sevastopolsky, Philip Grassal, Simon Giebenhain, ShahRukh Athar, Luisa Verdoliva, Matthias Nießner
3DV, 2025  
project page / video / arXiv / code

We learn to generate large displacements for parametric head models, such as long hair, with high level of detail. The displacements can be added to an arbitrary head for animation and semantic editing.

TriPlaneNet: An Encoder for EG3D Inversion
Ananta Raj Bhattarai, Matthias Nießner, Artem Sevastopolsky
WACV, 2024  
project page / video / arXiv / code

EG3D is a powerful {z, camera}->image generative model, but inverting EG3D (finding a corresponding z for a given image) is not always trivial. We propose a fully-convolutional encoder for EG3D based on the observation that predicting both z code and tri-planes is beneficial. TriPlaneNet also works for videos and in real time (check out the Live Demo).

How to Boost Face Recognition with StyleGAN?
Artem Sevastopolsky, Yury Malkov, Nikita Durasov, Luisa Verdoliva, Matthias Nießner
ICCV, 2023  
project page / video / arXiv / code & datasets

Can the power of generative models provide us with the face recognition on steroids? Random collections of faces + StyleGAN are the secret sauce. We release the collections themselves, as well as a new fairness-concerned testing benchmark.

Relightable 3D Head Portraits from a Smartphone Video
Artem Sevastopolsky, Savva Ignatiev, Gonzalo Ferrer, Evgeny Burnaev, Victor Lempitsky
arXiv, 2020  
project page / video / arXiv / talk video

By taking a simple selfie-like capture by a smartphone, one can easily create a relightable 3D head portrait. The system is based on Neural Point-Based Graphics.

media coverage: IEEE Spectrum (March '21), technology.org (January '21)
TRANSPR: Transparency Ray-Accumulating Neural 3D Scene Point Renderer
Maria Kolos*, Artem Sevastopolsky*, Victor Lempitsky
3DV, 2020  
project page / video / arXiv / code

An extension of Neural Point-Based Graphics that can render transparent objects, both synthetic and captured in-the-wild.

Neural point-based graphics
Kara-Ali Aliev, Artem Sevastopolsky, Maria Kolos, Dmitry Ulyanov, Victor Lempitsky
ECCV, 2020
project page / video / arXiv / code

Given RGB(D) images and a point cloud reconstruction of a scene, our neural network generates extreme novel views of the scene which look highly photoreal.

Coordinate-based texture inpainting for pose-guided human image generation
Artur Grigorev, Artem Sevastopolsky, Alexander Vakhitov, Victor Lempitsky
CVPR, 2019
project page / arXiv / supmat

How would I look in a different pose? Or in different clothes? A ConvNet with coordinate-based texture inpainting to the rescue.

Stack-U-Net: refinement network for improved optic disc and cup image segmentation
Artem Sevastopolsky, Stepan Drapak, Konstantin Kiselev, Blake M. Snyder, Jeremy D. Keenan, Anastasia Georgievskaya
SPIE Medical Imaging, 2019  
arXiv

An advanced version of "Optic disc and cup segmentation methods..." (see below), where a segmentation is performed by a U-Net stacked multiple times, and a validation is performed on large amounts of data provided by UCSF.

Optic disc and cup segmentation methods for glaucoma detection with modification of U-Net convolutional neural network
Artem Sevastopolsky,
Pattern Recognition & Image Analysis, 2017  
arXiv / code

Automatic segmentation of two organs on an eye fundus image allows medical doctors to make more accurate early diagnosis of glaucoma and evaluate its progression over time.



The webpage template was borrowed from the exciting page of Jon Barron.